T-Oil: A Contactless Edible Oil Adulteration Perception Using Terahertz Signals

Adulterated edible oil is a widespread issue due to limited production of high-cost oils, undermining fair competition in the food trade and posing health risks to consumers. Existing techniques for adulterated edible oil detection usually require trained professionals and specialized equipment, limiting their applicability in everyday situations. In this paper, we show that Terahertz (THz), a potential frequency band for emerging wireless communication technologies, offers molecular-level sensing capabilities. Leveraging this sensing capability, we propose T-Oil, a novel contactless system for detecting adulteration in edible oils. Specifically, T-Oil transmits a Terahertz signal to the sample being tested and analyzes changes in the absorption characteristics of the reflected signal to assess compositional differences, thereby identifying adulterated edible oils. In T-Oil, we design a two-stage deep learning model that employs contrast learning to identify the oil types present in an adulterated sample, and then apply a multi-source domain adaptation approach to quantify the proportions of the adulterants. With 25,950 signal samples collected, T-Oil can identify the oil types in the adulterated edible oil at an accuracy of 98.3% and quantify the detailed adulterated proportion at an accuracy of 96.4%.